The Agent Skills Directory

[COMMAND_EXECUTION]: The skill instructs the orchestrator to run date +%s to calculate a deadline and check the remaining time before each iteration. This is a functional and safe use of system commands for workflow logic control.
[INDIRECT_PROMPT_INJECTION]: The skill ingests a user-provided 'Topic' and 'Context' (such as codebase paths or documents) and passes this data to the Proposer and Critic subagents.
Ingestion points: External data from the Topic and Context inputs is interpolated into the prompts for subagents as described in SKILL.md.
Boundary markers: Absent. The ingested data is directly inserted into prompt templates without explicit delimiters or instructions to ignore embedded commands.
Capability inventory: The orchestrator has capabilities to execute the date command and perform file write/read operations in the /tmp directory (SKILL.md).
Sanitization: Absent. The skill does not perform validation or escaping of the ingested context before it is processed by subagents.
[SAFE]: The skill writes progress logs and discovered patterns to a file in the /tmp directory. This is used for session persistence and to allow the Proposer to avoid repeating previous ideas, which is standard behavior for managing temporary data.

ideate